Tuesday, July 14, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

Grok 4.5 inference costs 4.2x lower than Fable 5 and GPT-5.5 despite trailing on some benchmarks; efficiency analysis highlights cost-quality tradeoffs.

Cost-optimized frontier models gain narrative traction; hyperscalers and startups reassess inference spend versus latency/accuracy requirements.
Trade pressSlicast · July 9, 2026 · Global · Source: The Decoder
importance 58

xAI has released Grok 4.5, trained on tens of thousands of Nvidia GB300 GPUs for coding, agentic tasks, and knowledge work.

Benchmark results paint a mixed picture. On Terminal Bench 2.1, which tests complex command-line tasks, Grok 4.5 scores 83.3%, nearly matching GPT-5.5 at 83.4% and trailing Anthropic's Fable 5 by just one point at 84.3%. But the gaps widen elsewhere. On DeepSWE 1.1, which measures the ability to resolve real GitHub issues, Grok 4.5 reaches 53%, well behind GPT-5.5 at 67% and Fable 5 at 70%. On SWE Bench Pro, a curated set of harder software engineering problems, Grok 4.5 scores 64.7%, beating Opus 4.8 in some configurations though Opus 4.8 reaches 69.2% with maximum settings, while trailing Fable 5's 80.4%.

xAI says it relied on heavy data filtering, deduplication, and domain-specific selection during training to maintain data quality. The reinforcement learning phase covered hundreds of thousands of software-engineering tasks with automated scoring. xAI built the training infrastructure for asynchronous learning, allowing agentic runs to stretch over many hours while training continued in parallel.

Grok 4.5 costs $2 per million input tokens and $6 per million output tokens—far below the competition. Opus 4.8 costs $5 input and $25 output per million tokens; Fable 5 charges $10 input and $50 output per million; GPT-5.5 and GPT-5.6 sit at $5 input and $30 output per million.

xAI also says Grok 4.5 uses 4.2 times fewer tokens than Opus 4.8 on SWE Bench Pro tasks and delivers results at 80 tokens per second. Combined with lower per-token pricing, Grok 4.5 becomes by far the cheapest option in this performance tier, assuming the performance and efficiency gains hold up in practice.

The pricing strategy echoes what Chinese vendors like Zhipu and DeepSeek have been doing: get close enough on performance, then win on price.

Grok 4.5 is available now through Grok Build, Cursor, and the xAI console, with plugins live for Word, PowerPoint, and Excel. The model isn't available in the EU yet, with xAI targeting a mid-July launch. xAI trained Grok 4.5 alongside the code editor Cursor, which SpaceX acquired in mid-June for $60 billion in stock.

Read the original
Grok 4.5 inference costs 4.2x lower than Fable… · Slicast